Leaderboard

On-device LLM performance rankings powered by Glicko-2

iPhone 16 Pro Max

iOS

Rank

#9

Rating

1,966

±13 RD

Win Rate

95.0%

Conservative Rating

1,940

TG Rating

1,949

PP Rating

1,966

Matches

1,568

Record

1489W – 79L

Models Tested

ModelTG Median (tok/s)PP Median (tok/s)TG BestPP BestRuns
DeepSeek-R1-Distill-Qwen-7B-Q4_K_M1385.4978705.262767.97157404.572
tinygemma3-Q8_0324.4325507.49324.4325507.491
SmolLM2-135M-Instruct-Q8_0177.215002.24179.215020.622
Qwen3-0.6B-Q8_061.331240.4161.331240.411
DeepSeek-R1-Distill-Qwen-7B-Q4_K_L60.77587.8060.77587.801
Vintern-1B-v3_5-Q8_059.271473.7259.271473.721
DeepSeek-R1-Distill-Qwen-1.5B-IQ3_XS47.56514.0447.56514.041
gemma-3-1b-it.Q3_K_M45.6679.9345.6679.931
gemma-3-1b-it-Q8_040.61954.9040.61954.901
gemma-3-1b-it.Q8_040.48907.7840.63964.202
llama-3.2-1b-instruct-q8_032.6772.7939.03797.097
DeepSeek-R1-Distill-Qwen-1.5B-Q4_K_M28.5745.9728.5745.971
DeepSeek-R1-Distill-Qwen-1.5B-Q8_028.22526.8428.25529.162
DeepSeek-R1-Distill-Qwen-1.5B-Q4_K_M27.9844.5427.9844.541
qwen2.5-1.5b-instruct-q8_025.1852.4130.58586.273
Qwen2.5-Coder-3B-Instruct-abliterated-Q4_K_M24.34261.2324.34261.231
SmolLM2-1.7B-Instruct-Q8_024.18427.1624.39430.032
DeepSeek-R1-Distill-Qwen-1.5B-Q8_023.37155.2726.02287.073
Gemmasutra-Mini-2B-v1-Q6_K20.97287.4022.24346.637
Qwen2.5-Coder-3B-Instruct-abliterated-Q5_K_M20.68230.2820.68230.281
deepscaler-1.5b-preview-q8_020.5146.8020.5146.801
Qwen3.5-2B-IQ4_NL20.36340.5720.36340.571
gemma-3-4b-it-IQ4_NL19.28202.0719.28202.071
DeepSeek-R1-Distill-Qwen-1.5B-Q6_K_L19.27252.9233.94456.204
sonnet-llama-3.2-3b.Q4_K_M19.21172.2919.21172.291
gemma-3-4B-it-QAT-Q4_019.06213.0919.75215.652
bootes-qwen3_coder-reasoning-q4_k_m19.04195.5319.04195.531
Qwen3-4B-Instruct-2507-Q4_K_M18.93197.5818.93197.581
Hermes-3-Llama-3.2-3B.Q5_K_M18.80206.4818.80206.481
amoral-gemma3-4B-v2.IQ4_XS18.62195.4318.62195.431
LFM2-1.2B-F1618.5235.8118.5235.811
gemma-3n-E2B-it-IQ4_XS18.5029.6618.5029.661
Gemma-3-4B-VL-it-Gemini-Pro-Heretic-Uncensored-Thinking_Q4_k_m18.48198.2118.48198.211
gemma-3-4b-it-Q4_K_M18.10204.0618.10204.061
gemma-3-4b-it-q4_0_s17.66187.3719.79215.405
EXAONE-3.5-2.4B-Instruct-Q8_017.31263.2017.31263.201
DeepSeek-R1-Distill-Qwen-1.5B-Q8_016.5451.4616.5451.461
Qwen3-1.7B.Q6_K16.4830.0616.4830.061
DeepSeek-R1-Distill-Qwen-1.5B-f1615.75520.0515.75520.051
DeepSeek-R1-Distill-Qwen-1.5B-IQ2_M15.5822.2517.0423.322
Bootes-Qwen3_Coder-Reasoning.Q5_K_M15.20174.9815.20174.981
google_gemma-3-4b-it-Q6_K14.86199.1014.86199.101
llama-3.2-3b-instruct-q8_014.28238.3714.28238.371
gemma-3-4b-it.Q4_K_M14.27110.0318.02201.752
Phi-4-mini-reasoning-Q6_K14.23183.2214.23183.221
amoral-gemma3-4B-v2.Q4_K_M13.93158.5513.93158.551
gemma-3-4b-it-Q4_K_M13.76104.5017.88191.642
gemma-3n-E2B-it-Q4_113.5322.6613.5322.661
Llama-3.2-3B-Instruct-uncensored-Q8_013.46218.6514.08232.682
DeepSeek-R1-0528-Qwen3-8B-Q2_K13.2694.9413.2694.941

150 of 101 rows

1 / 3

Head-to-Head Record

150 of 353 rows

1 / 8

Performance by App Version

ImprovedRegressed

Compare With